Paraphrased from each maker’s own description. Select a line for the quote.
Checked for Gemini CLI, Hermes Agent and Pi, not yet for CodeBuddy Code: Price (Gemini CLI: ; Hermes Agent: ; Pi: ), Model providers (Gemini CLI: ; Hermes Agent: ,,,,,,SpaceXSI (Grok),Amazon Bedrock,,,Moonshot AI,,,,,,,,,Vercel AI Gateway,,,OpenCode Console (OpenCode Zen),Alibaba (Qwen),,,,Meta AI (Llama),,Microsoft Foundry (Azure AI Foundry),Nebius Token Factory (Nebius AI Studio),Novita AI,,,,,,Command Code,,,,,,,,,,, ; Pi: ,,,,,SpaceXSI (Grok),,,,,Z.ai,,,,,,,,,,OpenCode Console (OpenCode Zen),,,,Meta AI (Llama),,,Microsoft Foundry (Azure AI Foundry),, ), Bring your own key (Gemini CLI: ; Hermes Agent: ; Pi: )
Checked for Gemini CLI and Pi, not yet for CodeBuddy Code and Hermes Agent: Systems (Gemini CLI: ,, ; Pi: ,,, )
Not yet checked for any of these tools: agents it runs.
Provider names use the same listing as each tool page. Dotted values open their quote.
What each can do
Select an answer to see the maker’s sentence, or what we searched.
We have not checked CodeBuddy Code’s capabilities yet. 27 capabilities are checked for at least two of these tools. 9 get different answers, among them Approval modes, Subagents, Editor integration and Plan mode; 3 differ only in a limit. 7 others are checked for one tool only.
Editors, scripts and CI Editor integration, Headless mode, JSON output, GitHub integration, Commit attribution control
Review, steer and undo Plan mode, Read-only mode, Review diffs, Undo file edits
Parallel agents Parallel sessions, Git worktrees
Context and memory Project instruction files, Memory across sessions, Resume a session, Browse past sessions
Remote and mobile control Remote control, Steer from a phone, Notifications
Sandbox, permissions and privacy OS-level sandbox, Approval modes, Runs in a container, Telemetry opt-out
Images, web and your computer Image input, Web search, Fetch URLs, Browser automation, Computer use, Voice input
We have not checked these yet. They may or may not be supported.
,
11 capabilities not yet checked for Gemini CLI
Models, keys and usage Local models, Custom OpenAI-compatible endpoint
Editors, scripts and CI GitHub integration, Commit attribution control
Parallel agents Parallel sessions
Remote and mobile control Remote control, Steer from a phone
Sandbox, permissions and privacy Telemetry opt-out
Images, web and your computer Browser automation, Computer use, Voice input
We have not checked these yet. They may or may not be supported.
,
8 capabilities not yet checked for Hermes Agent
Editors, scripts and CI GitHub integration, Commit attribution control
Review, steer and undo Plan mode, Read-only mode
Sandbox, permissions and privacy OS-level sandbox, Telemetry opt-out
Images, web and your computer Fetch URLs, Computer use
We have not checked these yet. They may or may not be supported.
,
14 capabilities not yet checked for Pi
Editors, scripts and CI Editor integration, GitHub integration, Commit attribution control
Review, steer and undo Review diffs
Parallel agents Parallel sessions, Git worktrees
Context and memory Memory across sessions
Remote and mobile control Remote control, Steer from a phone
Images, web and your computer Web search, Fetch URLs, Browser automation, Computer use, Voice input
We have not checked these yet. They may or may not be supported.
3 capabilities are not yet checked for any of these tools and are not shown. Each answer is about the CLI agent in its column. Something the maker publishes separately, such as an editor plugin or a GitHub Action, shows as “With add-on”, even when it has its own listing here. How capabilities are checked
Votes can only raise a score: up to 20% of it reflects upvotes, graded on a curve against the most-voted tool. The square-root curve has a floor of 100 effective votes. Votes decay with a 90-day half-life. Maker self-votes do not count. Counts below 10 are hidden for display only. After activation, the score is 80% mindshare plus 20% of the greater of mindshare and the vote score. GitHub stars do not affect the Index score. Mindshare uses a log scale; votes use the square-root curve and a 100-vote floor. Ranks use unrounded scores. Mindshare and adoption bars are log-scaled; the others are linear within this comparison. How the score is built
—Votes can only raise a score: up to 20% of it reflects upvotes, graded on a curve against the most-voted tool. The square-root curve has a floor of 100 effective votes. Votes decay with a 90-day half-life. Maker self-votes do not count. Counts below 10 are hidden for display only.
—Votes can only raise a score: up to 20% of it reflects upvotes, graded on a curve against the most-voted tool. The square-root curve has a floor of 100 effective votes. Votes decay with a 90-day half-life. Maker self-votes do not count. Counts below 10 are hidden for display only.
—Votes can only raise a score: up to 20% of it reflects upvotes, graded on a curve against the most-voted tool. The square-root curve has a floor of 100 effective votes. Votes decay with a 90-day half-life. Maker self-votes do not count. Counts below 10 are hidden for display only.
—Votes can only raise a score: up to 20% of it reflects upvotes, graded on a curve against the most-voted tool. The square-root curve has a floor of 100 effective votes. Votes decay with a 90-day half-life. Maker self-votes do not count. Counts below 10 are hidden for display only.